Project Summary Candidate Goals and Objectives: With a background in Information Systems and Management, and Biostatistics, Dr. Zhang has demonstrated research records on electronic health record data mining to identify patterns of healthcare delivery that may be used to inform patient-centered and evidence-based healthcare. The proposal will provide additional training for Dr. Zhang on advanced machine learning, statistics, and evaluation methods in biomedical informatics for applications on clinical decision support (CDS). Dr. Zhang's long-term goal is to bringing innovation CDS development and evaluation through novel biomedical informatics and data science techniques. Institutional Environment and Career Development: Weill Cornell Medicine (WCM) provides ideal research facilities and training environment for Dr. Zhang. Dr. Jyotishman Pathak, Chief of Division of Health Informatics at Department of Health Policy and Research, will lead a multidisciplinary team of mentors: Drs. Jessica Ancker and Fei Wang at WCM, and Dr. Adam Wright at Harvard Medical School. Dr. Zhang also has collaborators in WCM and NewYork-Presbyterian Hospital who will support her in her training and research activities and provide clinical expertise. Research Aims Order sets are a type of CDS in computerized provider order entry (CPOE) to standardize decision making in the ordering process and encourage compliance with clinical practice guidelines. Previous literature on order set use has focused its effect on usability, workload, and physician satisfaction, but a knowledge gap remains with respect to the effect of order sets on care outcomes. The overall goal of the research study is to create a continuous improvement cycle for order sets with respect to a care outcome by rigorously learning from data. Aim 1 of the study will apply computational phenotyping and subtyping algorithms to identify cohorts of heart failure (HF) subtypes. Aim 2 will evaluate an existing order set intended for the care of HF patients on a care outcome defined as 30-day all-cause, unplanned readmission with a hypothesis that the use of this order set is associated with a better outcome. This will be achieved by building a range of outcome prediction models and evaluating the strength of each order set order as a predictor. Aim 3 will optimize the existing order sets using a metaheuristic optimization method such that its content collectively may have the largest positive effect on the outcome of 30-day all-cause unplanned readmission. The effects of order set use on the care outcome is measured using a causal inference technique in each iteration. The expected outcome is a framework to develop and evaluate HF order sets which may eventually be generalized to other clinical areas. Training from this proposal may lead to multi-site R01 studies of outcome-driven HF order sets and actual implementations.